Bits and bytes: the future of radiology lies in informatics and information technology


Advances in informatics and information technology are sure to alter the practice of medical imaging and image-guided therapies substantially over the next decade. Each element of the imaging continuum will be affected by substantial increases in computing capacity coincident with the seamless integration of digital technology into our society at large. This article focuses primarily on areas where this IT transformation is likely to have a profound effect on the practice of radiology.

Key points

Clinical decision support ensures consistent and appropriate resource utilization.

Big data enables correlation of health information across multiple domains.

Data mining advances the quality of medical decision-making.

Business analytics allow radiologists to maximize the benefits of imaging resources.

The rapid evolution of informatics and information technology promises to reshape the practice of radiology substantially over the next decade. Clinical decision support, big data, image mining, improvements in business analytics and Google and Apple in the healthcare space are leading to dramatic changes as relentless increases in computing power and memory capacity lead to deeper penetration of digital technology into radiologists’ daily lives.

Clinical decision support

Reducing variation in radiology practice is expected to improve population health management. Overutilization of imaging services can drive up healthcare costs and increase population health risk by causing needless subsequent evaluation of false positive and incidental findings; however, equally important, underutilization can also drive up healthcare costs and cause patient harm by leading to missed diagnoses and delayed treatments. The goal of clinical decision support (CDS) is to ensure consistent and appropriate resource utilization, thereby optimizing health benefits while reducing costs.

By providing appropriate use criteria and referral guidelines when imaging examinations are contemplated, robust CDS can guide referring practitioners, and even patients, to the appropriate imaging examination. It also promises to enable administrators to measure and adjust the capacity of advanced imaging resources based on usage patterns.

In the early 2000s, CDS for referring physicians was pioneered and piloted at Massachusetts General Hospital (MGH) and Brigham and Women’s Hospital (BWH) in Boston, Massachusetts, USA. The systems developed were shown to reduce inappropriate utilization of imaging resources [1, 2] and have become part of the culture at these institutions. On April 1, 2014, President Obama signed the “Protecting Access to Medicare Act of 2014,” which requires that physicians consult government-approved, evidence-based appropriate use criteria through a CDS system when ordering advanced diagnostic imaging exams (computed tomography [CT], magnetic resonance imaging [MRI], nuclear medicine and positron emission tomography [PET]). Currently, decision support is being used in over 150 health systems and over 1000 discrete acute care facilities in nearly every state [3]. The system developed at BWH was commercialized by Medicalis and is in use in numerous healthcare facilities throughout the United States. The system pioneered at MGH is based predominantly on the American College of Radiology’s (ACR’s) appropriateness criteria and has been adapted by the ACR in a product known as “ACR Select.” Both systems can be integrated with the electronic health records systems of major vendors, enabling seamless operability.

The European Society of Radiology has adapted ACR Select for European practice patterns in its product known as ‘ESR i-Guide’ ( A prototype of this system was piloted successfully in Barcelona, Spain, and is now being considered at many test sites throughout Europe.

CDS for radiologists can help reduce the variability of their descriptions and quantitation of imaging findings as well as the variability of conclusions and recommendations they draw from those findings. MGH has developed a decision support system for radiologists, which provides clinical guidance at the point of radiologic care through a user interface linked to the automated voice recognition system [4]. When natural language processing identifies a finding for which clinical guidance is available, a pop-up window appears that enables the user to input key imaging features. Suggested language for findings, impressions, and recommendations is built in real time, and the user can then insert this language into his or her report. This system has been shown to double the rate of compliance with accepted guidelines for follow-up recommendations consequent to incidental imaging findings [5]. Next-generation prototypes of this system include images that inform the radiologist of pathologic subtypes for various selected attributes, enabling the radiologist to match the case in hand with reference images for each imaging feature; suggested report language is built in real time as the radiologist selects the images that best correspond to those at hand.

In addition to reducing variations in recommendations for clinically significant findings, decision support systems for radiologists could lead to more trackable conclusions and recommendations, enabling the creation of automated workflows for critical results reporting and follow-up of recommended diagnostic testing.

The University of California at Los Angeles (UCLA) has created linked information systems that compare radiologic diagnoses with pathologic or surgical diagnoses to generate integrated, interdisciplinary, and congruent reports. Such accurate, integrated, and actionable reports are already in a clinical use within UCLA’s electronic medical record (EPIC). We must integrate our radiologic information, relevant pathologic (molecular and genomic) information and clinical information as provided by modern electronic health record systems. UCLA’S “Integrated Diagnostic” program created a reporting tool which generates correlation of imaging and pathologic data streams to provide integration whose strength lies in explaining agreement or resolving incongruence between radiologic and pathologic findings in a single report [6]. Data-driven diagnostic decision support based on such integrated reports will increase productivity by accelerating time to accurate diagnosis and by reducing time costs, particularly if the integration encompasses therapeutic procedures and activities. There is a strong rationale for diagnostic and interventional radiology to remain and grow together.

The impact of big data

Digital data is projected to reach 35 zettabytes by 2020, a 44-fold increase from 2009 [7]. A 2011 McKinsey report estimated that the health care industry can potentially realize $300 billion in annual value by leveraging big data [8]. “Big data goes beyond size and volume to encompass such characteristics as variety, velocity, and with respect specifically to health care, veracity” [8]. Volume refers to the scale of the data, variety refers to the degree to which the data is structured or unstructured, velocity refers to the speed at which data is produced and collected, and veracity refers to the data quality certainty. The “big” part of big data refers to volume, variety and velocity. So-called “big data” sets are those so vast or complicated that their analysis and management require “massively parallel software running on tens, hundreds, or even thousands of servers” [9]. Big data in health care encompasses a wide range of domains including genomics, proteomics, phenotype information, and the electronic health record and medical imaging, inclusive of radiology, pathology, cytology, and laboratory medicine. These components will continue to change over time. The value of big data cannot be overestimated. With big data, it is possible to correlate health information such as similar diagnoses, findings, genetics, clinical presentations, therapeutic responses, outcomes, and prognoses. The implications of big data for medical imaging are myriad. Taking lessons from the oil industry, space exploration, and military applications, scientists are applying machine learning to analyze imaging data in detail and correlate it with other medical data. The use of machine learning promises better decision support for medical imaging, improved quantitative imaging, improved computer-assisted diagnoses, improved computer-assisted radiology for structured report generation, precision in diagnosis, and real-time correlation with other medical data. It also promises to enable true outcome evaluations for medical imaging and the bridging of molecular imaging and other diagnostics to the clinical arena. Potential pitfalls in the use of big data relate to data quality, difficulty in identifying patients, security and management of protected health information, and the ease of use of big data.

Users must be trained in big data management, emphasizing the patient-specific nature of the data. Therefore, it is important that data be itemized in a reversible fashion for research purposes. The data must be verified and dated with the identification of the responsible “owner,” and it must be carefully defined and precisely formatted. Individuals should not be allowed access to big data without careful training and oversight; audit trails of access modifications and uses must be maintained.

We must be careful not to assume a causal relationship when statistically significant correlations are found. Further research is needed to establish a cause and effect between similarly dispersed variables.

The quality of data contained in big data sets must be assessed and maximized. The electronic health record is replete with inaccurate information, free text, conjecture, assumptions, and as-yet unproven diagnoses. The health care vocabulary is imprecise; many terms often considered synonymous, in fact, have definitions that merely overlap. The development of a radiology imaging ontology will prove helpful. Patients have not been identified the same way across health systems, and electronic health records are also not uniformly used across health care systems or among health care practitioners. Thus, it is difficult to assume population statistics when examining big data sets.

Tools are necessary to automate human interaction with big data sets. For instance, natural language processing (NLP), a field that combines computer science and linguistics, provides tools for representing natural language so it can be used for computation [10]. Another branch of computer science, deep machine learning, employs multiple processing layers to model multiple levels of abstraction and discover intricate structures in large data sets [11].

What does image mining offer?

Data mining is one step in a process of knowledge discovery in databases. In radiology, image data is the sum of the medical imagery and the metadata that accompanies it in our electronic health records. Data mining techniques include grouping of similar data objects, decision trees (tree-like graphs or models of decisions), neural networks (systems that derive meaning from complicated or imprecise data), and evolutionary algorithms (heuristic search algorithms that mimic natural selection).

Data mining offers the possibility of using references and standards derived from a trusted cohort to improve the quality of medical decision-making. For example, a CT dose index registry clusters dose data from patients with known characteristics. The cohort can be filtered by selected characteristics, enabling analysis of cohort features and outcomes in comparison to an index patient or patient population. At the University of Wisconsin, a project executed in partnership with General Electric has enabled distribution of University of Wisconsin CT protocols on GE CT scanners. Over 200 CT protocols per CT platform have been deployed by organ and clinical indication for 3 patient sizes. These have resulted in over 50,000 independent measures of dose and image quality, one for each patient examination. Data mining facilitates the comparison of patient-specific radiation doses to doses in the cohort of similarly sized patients undergoing similar procedures in the large database—i.e., to meaningful reference data.

A second example of cohort comparison can be found in systems designed to monitor the response to specific therapies. Databases that track imaging findings in response to therapeutic interventions can be mined to compare the responses of specific patients to others within a cohort of patients with a similar clinical condition and treatment profile. This can assist in outcome prediction for the given clinical scenario and treatment algorithm.

Data mining combined with radiomics, which is an emerging area within radiology in which images are converted into mineable data and then correlated with genomic, clinical and other data sets for decision support [12], offers the option to discover new imaging features not detectable through human observation. Radiomics is currently being used most commonly in oncology to potentially aid cancer detection, diagnosis, assessment of prognosis, prediction of response to treatment, and monitoring of disease status [12]. Pitfalls in data mining relate to the quality of the data within the data sets at hand. Bad data leads to pollution of subsequent analyses. While IBM has purchased Merge to provide roughly 30 billion images for Watson to digest, the quality of this data set will affect the downstream success of deep learning initiatives using this super-computing technology. To reach the practical clinical level, something less than a super-computer will be used, i.e., cloud computing.

Data mining will improve the quality of care we provide. However, it is critical that we take ownership of the information that is available so our expertise in medical imaging is leveraged to its best advantage.

Opportunities for business analytics

Economic pressures in healthcare have continued to raise the importance of business analytics—also known as “business intelligence”—for managing our practices [13]. The radiology value chain extends from the decision to pursue an imaging examination to the prescription of a treatment based on imaging findings. As radiology is a business with high fixed costs owing to the high price of imaging equipment, we must maintain high capacity utilization to ensure that we extract the greatest benefit from our imaging resources, machines, personnel, and accoutrements. Business analytics enable capacity analyses to ensure that the number of radiologists assigned to an imaging service is of the right size to manage the workload. Moreover, business analytics can help insure that the skill levels of radiologists and associated professional staff are matched to the clinical problems at hand.

A basic principle in economics is that productivity growth is a must for raising living standards. Similarly, growth in radiologists’ productivity has contributed to a rise in the contribution margin for our imaging services, defined as revenue net of direct expenses. By providing metrics of variables such as diagnostic accuracy, turnaround time, scanner utilization, and pre- and post-procedural wait times, business analytics enable us to better manage not only the productivity of individual radiologists, but also the productivity of the entire value chain [13]. Radiologists are only part of that chain, but an important part. Their accuracy is pivotal.

The impact of Apple and Google healthcare entries

While currently not specific to radiology, there are other technology trends that could have a substantial impact on how we practice medicine. According to Bill Maris, President and Managing Partner of Google Ventures, “Right now is the ‘transistor moment’ for the human body. “In the coming decades, health care will begin to improve at the same radical pace we’ve seen in computing” [14]. The areas of information technology (IT) innovation most relevant to healthcare include artificial intelligence, the Internet of Things (IoT), “small data,” and the harnessing of big data.

Artificial intelligence

Artificial intelligence can be divided into artificial narrow intelligence (ANI) and artificial general intelligence (AGI). ANI refers to specific tasks that computers do as well as or better than humans. Google’s self-driving cars are one example. As of April 2016, after 6 years and more than 1.5 million miles of testing, they had had a total of just 21 accidents—20 of them due to human error [15, 16]. Another example can be found in IBM’s Watson, when it played Jeopardy and beat two champions in 2011 [17].

AGI refers to computers that go beyond learning specific tasks to performing higher-order syntheses. When computers start emulating thought processes, AGI is at play. Computer science pioneer Allen Turing proposed that if one third of the time someone could not distinguish between a human and a computer in conversation, then AGI would have been achieved. This test, often called the “Turing test,” was passed recently when scientists at Princeton artificial intelligence laboratories created a so-called chatterbot, a computer program designed to simulate the conversation of a human being; one third of the time, while communicating with the chatterbot in an instant messaging format, humans believed that they were conversing with another human. More recently, the definition of AGI has been reframed as the ability of a computer to do the things humans do without “thinking”. In Scientific American, a new test of AGI and “machine consciousness” was proposed [18], whereby a computer would be considered to demonstrate AGI if it could identify “what’s wrong with this picture” when examining images with incongruent components.

The Internet of Things (IoT) and “small data”

The IoT refers to the connectivity and interoperability of increasingly smart objects, such as sensors, controllers, wearables, and other technologies, including an increasing array of medical devices. It provides the ability to accumulate information from multiple sources to automate and improve/optimize processes and decision-making. The IoT is driven by advances in microelectronics, power management, wireless technologies, and advanced software approaches [19]. An important principle of the IoT is that the utility of the captured data is inversely proportional to its latency. For example, devices that enable remote physiological monitoring are nearly useless if the transmitted data is not received and understood in real time. The number of connected devices per person is increasing exponentially. Having gone from 1.84 in 2010 to 3.47 in 2015, it is expected to reach 6.58 by 2020 [20]. At the same time, technology adoption is increasing exponentially [21].

Apple has invested heavily in healthcare IT with its Health Kit technology. Partnering with Epic Systems and the Mayo Clinic, Health Kit is a software framework included in i0S8 that can collect, store, and present health information from applications designed to communicate with it. These applications are often associated with wearable sensors for capturing health or fitness data. Such data will provide many benefits, including the opportunity to better predict the pretest probability for certain diseases and thus increase the diagnostic accuracy of our imaging examinations. This is the kind of “small data” growth that can inform health and wellness well into the future.


As stated by William Gibson, “the future is already here, it’s just not very evenly distributed” [22]. Stated another way, the future is here and it is our job to embrace it. Over the next 10 years, we are likely to see augmented human intelligence, and our roles as diagnosticians will change. We need to be using informatics, IT, the IoT, and the advantages of big data to ensure that we will continue to offer added value to patients and the greater medical community.


  1. 1.

    Sistrom CL, Dang PA, Weilburg JB, Dreyer KJ, Rosenthal DI, Thrall JH (2009) Effect of computerized order entry with integrated decision support on the growth of outpatient procedure volumes: seven year time series analysis. Radiology 251:147–155

    Article  PubMed  Google Scholar 

  2. 2.

    Ip IK, Schneider L, Seltzer S et al (2013) Impact of provider-led, technology-enabled radiology management program on imaging. Am J Med 126:687–692

    Article  PubMed  Google Scholar 

  3. 3.

    Personal communication, Robert Cooke, National Decision Support Company, on October 12, 2016

  4. 4.

    Boland GWL, Thrall JH, Gazelle GS et al (2011) Decision support for radiologist report recommendations. JACR 8:819–823

    PubMed  Google Scholar 

  5. 5.

    Lu MT, Rosman DA, Wu CC et al (2016) Radiologist point-of-care clinical decision support and adherence to guidelines for incidental lung nodules. JACR 13:156–162

    PubMed  Google Scholar 

  6. 6.

    O’Reilly KB (2015) Integrating AP and radiology, inch by inch. CAP Today. Accessed 29 June 2016

  7. 7.

    Gantz J, Reinsel D (2010) The digital universe decade – are you ready? Accessed 30 April 2016)

  8. 8.

    Institute for Health Technology Transformation. Transforming health care through big data strategies for leveraging big data in the health care industry (

  9. 9.

    Jacobs A (2009) The pathologies of big data. ACM Queue 7:1–12

    Article  Google Scholar 

  10. 10.

    Yim WW, Yetisgen M, Harris WP, Kwan SW (2016) Natural language processing in oncology: a review. JAMA Oncol 2:797–804

    Article  PubMed  Google Scholar 

  11. 11.

    LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521:436–444

    CAS  Article  PubMed  Google Scholar 

  12. 12.

    Gillies RJ, Kinahan PE, Hricak H (2016) Radiomics: images are more than pictures, they are data. Radiology 278:563–577

    Article  PubMed  Google Scholar 

  13. 13.

    Cook TS, Nagy P (2014) Business intelligence for the radiologist: making your data work for you. J Am Coll Radiol 11:1238–1240

    Article  PubMed  Google Scholar 

  14. 14.

    Maris B. Medicine’s transistor moment: 8 emerging technologies that could revolutionize the life sciences. Google Ventures Library Web site. Published May 26, 2015. Accessed 23 June 2016

  15. 15.

    Davies A (2016) Google’s self-driving car caused its first crash. Wired. Accessed 23 June 2016

  16. 16.

    Google self-driving car project Web site. Monthly reports. Accessed 23 June 2016

  17. 17.

    Best J (2013) IBM Watson: The inside story of how the Jeopardy-winning supercomputer was born, and what it wants to do next. TechRepublic. Accessed 23 June 2016

  18. 18.

    Koch C, Tononi G (2011) A test for consciousness. Sci Am 304:44–47

    Article  PubMed  Google Scholar 

  19. 19.

    The internet of things (IoT) in health care. Research brief, The Advisory Board Company. April 8, 2015

  20. 20.

    Evans D (2011) The internet of things: how the next evolution of the internet is changing everything. Cisco Internet Business Solutions Group White Paper

  21. 21.

    Desilver D (2014) Chart of the Week: The ever-accelerating rate of technology adoption. PewResearchCenter Web site. Accessed 23 June 2016

  22. 22.

    Broadband Blues (2001) The Economist. Accessed 23 June 2016

Download references


We would like to thank members of the International Society for Strategic Studies in Radiology (IS3R) for contributing to the discussion of the above topics. We would also like to thank Hedi Hricak for her review and helpful comments and Ada Muellner for her editing. The scientific guarantor of this publication is James A. Brink, MD. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. The authors state that this work has not received any funding. No complex statistical methods were necessary for this paper. Institutional review board approval was not required because no research was conducted.

Presented at the biannual meeting of: International Society of Strategic Studies in Radiology, Amsterdam, The Netherlands, August 28, 2015.

Author information



Corresponding author

Correspondence to James A. Brink.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Brink, J.A., Arenson, R.L., Grist, T.M. et al. Bits and bytes: the future of radiology lies in informatics and information technology. Eur Radiol 27, 3647–3651 (2017).

Download citation


  • Medical informatics
  • Information technology
  • Clinical decision support
  • Data mining
  • Artificial intelligence